Autonomous surrogate model creation platform

Abstract

According to some embodiments, a surrogate model creation computer system includes a user interface to interact with a subject matter expert to create a scripted physics-based model workflow associated with an industrial asset. The surrogate model creation computer system further includes a surrogate model creation engine to automatically execute the scripted physics-based model workflow, in connection with a physics-based model of the industrial asset, to generate at least one response surface of the physics-based model workflow. The surrogate model creation engine may then use the response surface and a machine learning process to automatically create a surrogate model of the industrial asset. It may then be arranged to output the surrogate model for use by a substantially real-time analytics package associated with the industrial asset. According to some embodiments, scripting for a physics-based model workflow may also be automatically created.

The invention claimed is: 
     
       1. A surrogate model creation computer system, comprising:
 a user interface operative to interact with a subject matter expert via a subject matter device, wherein an output of the interaction is a scripted physics-based model workflow associated with an industrial asset, the scripted physics-based model workflow comprising a series of automatically performable steps adapted to create a trained surrogate digital model of the industrial asset, wherein the trained surrogate digital model is a simulation model of the industrial asset; and 
 a surrogate model creation engine, coupled to the user interface, adapted to:
 automatically execute the scripted physics-based model workflow, in connection with a physics-based model of the industrial asset, to generate at least one response surface of the scripted physics-based model workflow, the response surface including a plurality of solution points; 
 train a surrogate digital model of the industrial asset using the response surface and a machine learning process to automatically create the trained surrogate digital model of the industrial asset; 
 arrange to output the trained surrogate digital model for use by a substantially real-time analytics package associated with the industrial asset; 
 constantly, via one or more runners, scan one or more databases of field product data for updated field data associated with the use of the trained surrogate digital model by the substantially real-time analytics package; 
 automatically receive from the one or more databases of field product data, updated field data; 
 automatically determine, by the runners, three or more downstream files that exist downstream of the updated databases, wherein a first of the three or more downstream files include the scripted physics-based model workflow, a second of the three or more downstream files include the response surface generated via execution of the scripted physics-based model workflow in connection with the physics-based model of the industrial asset and a third of the three or more downstream files include the trained surrogate model; 
 automatically update the downstream files with the updated field data; 
 determine, by the system, the updated field data is outside a range of input parameters used to generate the response surface; 
 automatically generate a collection of new operating points; 
 and 
 automatically re-train the trained surrogate digital model with automatically generated collection of new operating points. 
 
 
     
     
       2. The system of  claim 1 , further comprising:
 a repository data store to contain the scripted physics-based model workflow for subsequent access and modification. 
 
     
     
       3. The system of  claim 2 , wherein the repository data store further contains at least one of: (i) metadata about the scripted physics-based model workflow, (ii) an industrial asset identifier, (iii) a user identifier, (iv) a software tool identifier, (v) an operating point, (vi) a range of operating points, (vii) a release date, (viii) a release version, and (ix) uncertainty data. 
     
     
       4. The system of  claim 3 , wherein the repository data store is a searchable database adapted to be accessed and updated by subject matter experts. 
     
     
       5. The system of  claim 1 , wherein the surrogate model creation engine is further configured to:
 receive subsequent audit information associated with operation of the industrial asset, and 
 update the trained surrogate digital model based on an accuracy assessment of the audit information. 
 
     
     
       6. The system of  claim 1 , wherein at least one of the execution of the scripted physics-based model workflow and an execution of the machine learning process are associated with at least one of: (i) a high-performance computing center, and (ii) a cloud-based computing environment. 
     
     
       7. The system of  claim 1 , wherein the scripted physics-based model workflow is associated with a plurality of high-fidelity physics-based models. 
     
     
       8. The system of  claim 1 , wherein the scripted physics-based model workflow is associated with at least one of: (i) a component level model, (ii) a module level model, (iii) a system level module, (iv) data that varies in space, and (v) data that varies in time, (vi) input parameters, and (vii) post-processing. 
     
     
       9. The system of  claim 1 , wherein the substantially real-time analytics package is associated with at least one of: (i) an engineering control application, (ii) an edge computing application, (iii) inspection analytics for manufacturing quality assurance, and (iv) multi-disciplinary design optimization. 
     
     
       10. The system of  claim 1 , wherein the surrogate model is associated with a defined space of solution points. 
     
     
       11. The system of  claim 1 , wherein training the surrogate digital model of the industrial asset further comprises the surrogate model creation engine generating a first collection of training Design Of Experiment (“DOE”) operating points and a first collection of validation DOE operating points different than the training DOE operating points; and
 wherein re-training the trained surrogate digital model of the industrial asset further comprises generating a second collection of training DOE operating points. 
 
     
     
       12. The system of  claim 1  further comprising:
 a license control key configured to allocate components of the scripted physics-based model workflow to a cloud computing environment in accordance with software licenses associated with each component and an optimization routine. 
 
     
     
       13. A method associated with a surrogate model creation computer system, comprising:
 interacting with a subject matter expert via a user interface of a subject matter device, wherein an output of the interaction is a scripted physics-based model workflow associated with an industrial asset, the scripted physics-based model workflow comprising a series of automatically performable steps adapted to create a trained surrogate digital model of the industrial asset, wherein the trained surrogate digital model is a simulation model of the industrial asset; 
 automatically executing, by a surrogate model creation engine, the scripted physics-based model workflow, in connection with a physics-based model of the industrial asset, to generate at least one response surface of the scripted physics-based model workflow, the response surface including a plurality of solution points; 
 training a surrogate digital model of the industrial asset using the response surface and a machine learning process to automatically create the trained surrogate digital model of the industrial asset; 
 arranging to output the trained surrogate digital model to be used by a substantially real-time analytics package associated with the industrial asset; 
 constantly, via one or more runners, scanning one or more databases of field product data for updated field data associated with the use of the trained surrogate digital model by the substantially real-time analytics package; 
 automatically receiving from the one or more databases of field product data, updated field data; 
 automatically determining, by the runners, three or more downstream files that exist downstream of the updated databases, wherein a first of the three or more downstream files include the scripted physics-based model workflow, a second of the three of more downstream files include the response surface generated via execution of the scripted physics-based model workflow in connection with the physics-based model of the industrial asset and a third of the three or more downstream files include the trained surrogate model; 
 automatically updating the downstream files with the updated field data; 
 determining, by the system, the updated field data is outside a range of input parameters used to generate the response surface; 
 automatically generating a collection of new operating points; 
 and 
 automatically re-training the trained surrogate digital model with the automatically generated collection of new operating points. 
 
     
     
       14. The method of  claim 13 , further comprising:
 placing the scripted physics-based model workflow into a repository data store for subsequent access and modification. 
 
     
     
       15. The method of  claim 13 , further comprising:
 receiving subsequent audit information associated with operation of the industrial asset; and 
 updating the trained surrogate digital model based on an accuracy assessment of the audit information. 
 
     
     
       16. A non-transitory, computer-readable medium storing instructions that, when executed by a computer processor, cause the computer processor to perform a method associated with a surrogate model creation computer system, the method comprising:
 interacting with a subject matter expert via a user interface of a subject matter device, wherein an output of the interaction is a scripted physics-based model workflow associated with an industrial asset, the scripted physics-based model workflow comprising a series of automatically performable steps adapted to create a trained surrogate digital model of the industrial asset, wherein the trained surrogate digital model is a simulation model of the industrial asset; 
 automatically executing, by a surrogate model creation engine, the scripted physics-based model workflow, in connection with a physics-based model of the industrial asset, to generate at least one response surface of the scripted physics-based model workflow, the response surface including a plurality of solution points; 
 training a surrogate digital model of the industrial asset using the response surface and a machine learning process to automatically create the trained surrogate digital model of the industrial asset; 
 validating the trained surrogate digital model against the scripted physics-based model workflow; 
 arranging to output the trained surrogate digital model to be used by a substantially real-time analytics package associated with the industrial asset; 
 constantly, via one or more runners, scanning one or more databases of field product data for updated field data associated with the use of the trained surrogate digital model by the substantially real-time analytics package; 
 automatically receiving from the one or more databases of field product data, updated field data; 
 automatically determining, by the runners, three or more downstream files that exist downstream of the updated databases, wherein a first of the three or more downstream files include the scripted physics-based model workflow, a second of the three or more downstream files include the response surface generated via execution of the scripted physics-based model workflow in connection with the physics-based model of the industrial asset and a third of the three or more downstream files include the trained surrogate model; 
 automatically updating the downstream files with the updated field data; 
 determining, by the system, the updated field data is outside a range of input parameters used to generate the response surface; 
 automatically generating a collection of new operating points; 
 and 
 automatically re-training the trained surrogate digital model with the automatically generated collection of new operating points. 
 
     
     
       17. The medium of  claim 16 , wherein the scripted physics-based model workflow is associated with a plurality of high-fidelity physics-based models. 
     
     
       18. The medium of  claim 16 , wherein the substantially real-time analytics package is associated with at least one of: (i) an engineering control application, (ii) an edge computing application, (iii) inspection analytics for manufacturing quality assurance, and (iv) multi-disciplinary design optimization. 
     
     
       19. The medium of  claim 16 , wherein the trained surrogate digital model is associated with a defined space of solution points and the surrogate model creation engine creates a collection of training Design Of Experiment (“DOE”) operating points and a collection of validation DOE operating points different than the training DOE operating points. 
     
     
       20. The system of  claim 1 , wherein metadata about the scripted physics-based model workflow including at least one operating point is updated with updated field data.